Detecting psychosis via natural language processing of social media posts: potentials and pitfalls

The early detection and continuous monitoring of psychosis is of utmost importance in ensuring timely and effective treatment. Current mental health care is unable to meet this demand, partially because methods to detect psychosis are relatively time-intensive and not scalable to large populations. Consequently, there has been an increasing focus on the potential of passive data collection from digital devices to overcome this issue. In the present article, we explore whether the analysis of social media (SM) posts through natural language processing (NLP) could improve the detection of psychosis. We first demonstrate how freely expressed speech can be processed automatically in the laboratory to predict and classify psychosis with high levels of accuracy. We further outline the current state of psychosis classification from SM-derived data and discuss methodological issues that are hampering progress in this field. Finally, we delve into potential pitfalls of such systems and provide insight into how these may be circumvented.